Stock Market Behavior Predicted by Rat Neurons

Page created by Adam Potter
 
CONTINUE READING
Stock Market Behavior Predicted by Rat Neurons
Stock Market Behavior Predicted by Rat
  Neurons
              by Timothy C. Marzullo, Neuroscience Program, University of Michigan, Ann Arbor
              Edward G. Rantze, Red Antze, Inc., Cumming, Georgia
              Gregory J. Gage, Biomedical Engineering, University of Michigan, Ann Arbor

  We here report for the first time, to the best of our knowledge, rat motor cortex neurons predicting the behavior of the
  American stock market. We implanted the motor cortex of the brains of rats with silicon electrodes. Using the correlation
  technique, we monitored the activity of neurons in our rats while simultaneously tracking the activity of stocks in the U.S.
  stock market.

  Background: Hedge Funds
  Hedge funds burgeoned in the early 1990's as a popular alternative to the conventional, and
  more regulated, mutual funds. Hedge funds have often used alternative methods, such as             Figure 1 (top): Behavioral
  various human social factors, to predict future performance of the stock market. However, we            apparatus: rat trained
  here propose an alternative alternative method.                                                           on a brain-machine
                                                                                                     interface task while stocks
  Methods: Correlation Analysis                                                                         simultaneously tracked.
  For nine days, neural activity in the form of firing rates (which are the number of electrical
  discharges per second) from recorded neurons (n=94) of three rats were averaged each day as

22 | Annals of Improbable Research | July-August 2006                                                           www.improbable.com
the rats learned to use a brain-machine interface1 to obtain food pellets.
  Mean firing rate data per day were stored using custom software (MATLAB, Mathworks Inc., Natick, MA), along with
  the closing stock prices for the same day for all corporations listed on NASDAQ, the New York Stock Exchange, and the
  American Stock Exchange (n=4195). Correlation coefficients were obtained using the corrcoef function of MATLAB, and
  only stocks that had significant coefficients (p f d-1       a short                                                           (1)
                         fd < f d-1       abuy                                                              (2)
                         fd ≈ f d-1       ahold                                                             (3)
  where fd-1 is the firing rate from day d - 1 and a is the action taken, a = {buy; short; hold}. Stated simply, if the rats’
  neurons increased firing rates, we would simulate a “short” of the stock; if the firing rates decreased, we would “buy” the
  stock. If no change occurred (± 1 impulse/s), we did not trade that day (hold). To determine the success of our predictions,
  the actual value of the stock was observed on day d+1, and we calculated our profits and losses. Brokerage fees were not
  included in this analysis.

                                                    Coca-Cola Bottling Co. (COKE) Stock (US$)
                                                                                                58.5                                                     6
  Results
  We found that 74 stocks were responsive
  to the firing rates of our rats. Figure                                                        58                                                      5.5
  2 shows an example of one stock

                                                                                                                                                               Firing Rate (Imp/s)
  (COKE, Coca-Cola Bottling Company                                                             57.5                                                     5
  Consolidated) that was positively
  correlated with the rat neurons. Table 1
                                                                                                 57                                                      4.5
  groups the responsive stocks by sector.
  Though interesting clusters emerge in
  the financial and technology industries,                                                      56.5                                                     4
  the theoretical implications are beyond
  the scope of this paper.
                                                                                                 56                                                      3.5
  In our prediction experiments, we found
  a similar number of stocks that responded
                                                    55.5                                                                     3
  to a lag of one day (n=68). Figure 3               June-10 June-14 June-15 June-16 June-17 June-18 June-21 June-23 June-24
  shows the output of the stock trading
  simulation for one exemplar example                           Figure 2: Coca-Cola Stock Price (red) and average firing rates of
  stock (ASFI, Asta Funding, Inc.). Figure 3A indicates     neurons (blue) from rat motor cortex over 9 days in 2004. Correlation
  the results of the predictions, while Figure 3B shows                                                        coefficient = 0.704.
  our return on investment using the directives provided
  by the contrarian predictive model.

  Discussion
  For our analysis, we adopted the standard practice in neurophysiology where researchers will record a population of
  neurons, say 500, and find 50 that respond to a certain stimulus. The researchers will then decide to focus on the cells that
  showed responses and subject these to further statistical analysis. Thus, based on the work of our colleagues, we believe
  our methods are sound.
  We found that stocks correlate with the firing rates of motor cortex neurons in rats. We also generalized our model to
  predict future stock price, and we made $435 from an initial $1000 investment in 20 days by using neuronal firing rates to
  predict whether to buy, short, or hold shares in Asta Funding, Inc.

www.improbable.com                                                                                     July-August 2006 | Annals of Improbable Research | 23
Conclusion
                                  Nobel Prize-winning economist Paul Samuelson said in a 1967 declaration to the U.S. Senate that buying a mutual fund is
                                  worse than throwing darts at a dartboard. As a consequence, index and hedge funds are now popular. We say that if you are
                                  not using a rat motor cortex model of stock price, you might as well be using a mutual fund.

  A                             Buy
                                                                                                                  Appendectal Discussion

      Model Prediction
                                Hold
                                                                                                                  We are on the verge of a paradigm shift we call the Gage / Rantze /
                                                                                                                  Marzullo (GRM, or the Generalized Revenue Model) Motor Cortex
                               Short
                                                                                                                  Rattus norvegicus Theory of Societal Urges. The neurons of our rats
                                       1   2   3   4   5   6   7   8   9   10 11 12 13 14 15 16 17 18 19 20
  B                            1450
                                                                                                                  are in some mysterious way tied to humans’ purchase patterns which
                               1400                                                                               ultimately manifest as fluctuations in the American Stock Market.
                               1350

                               1300
                                                                                                                 The Gaia hypothesis, proposed by James Lovelock in the 1960’s, states

      Portfolio Value (US $)
                               1250                                                                               the Earth entire is a living organism.3 The data presented here are
                                                                                                Final Value
                                           Initial Investment
                               1200
                                                 ($1000)
                                                                                                  ($1,435)        consistent with this theory. We are all tied in a great circle of life,4
                               1150                                                            +43% Increase
                               1100
                                                                                                                  where our hopes, dreams, aspirations, triumphs, despairs, boredoms, and
                               1050
                                                                                                                  loves are inextricably linked to the creatures of the Earth. Research in
                               1000                                                                               1934 proved that the solar cycles of 1929 were correlated to the closing
                                950
                                       1   2   3   4   5   6   7   8   9 10 11 12 13 14 15 16 17 18 19 20         stock prices of the London and New York stock exchanges of the same
                                                                        Trading Day
                                                                                                                  year.5 Though we do not have access to rat motor cortex firing rates from
                                                                                                                  19296, our future experiments will do a triple correlation between rat
Figure 3: Results of predicting closing stock price
                                                                                                               motor cortex firing rates, the American and London Stock Markets, and the
of ASFI on day d + 1 from average firing rates on
                                                                                                               2006 solar radiation flux.
day d. A. Output of contrarian prediction model.
B. Simulation of US $1000 investment using trade                                  We focused on rats in this study, but we would not be surprised if the stock
information obtained from predictions.                                            market was correlated to the behavior of American White House squirrels,
                                                                                  Jamaican fruit bats, Tasmanian devils, and New England codfish. As a final
                                  note, we wonder what would happen to the stock market should species become extinct. Given Earth’s current global
                                  biodiversity crash and mass extinction crisis,7 future human economic success may be neither assumed nor assured.

                                  Notes
                                  Results from the study were previously presented at the 2005 annual Society for Neuroscience meeting in Washington,
                                  D.C.
                                  Conflict of Interest Statement: The authors of this study do not personally own any stocks in Asta Funding or Coca-Cola,
                                  unless one includes index funds that represent the whole stock market.

                                  References
                                  1. Brain-machine interfaces are devices that are controlled by the self-modulation of brain activity. The rat data presented
                                  here were acquired as part of a broad experiment examining brain-machine interface algorithm designs.
                                  “Naive Coadaptive Cortical Control,” Gregory J. Gage, Kip A. Ludwig, Kevin J. Otto, Edward I. Ionides, and Daryl R.
                                  Kipke, Journal of Neural Engineering, vol. 2, no. 2, 2005, pp. 52-63.
                                  2. “Profitability of Short-term Contrarian Strategies: Implications for Market Efficiency,” Jennifer Conrad, Mustafa N.
                                  Gultekin, and Gautam Kaul, Journal of Business Economic Statistics, vol. 15, no. 3, 1997, pp. 379-86.
                                  3. Gaia: A New Look at Life on Earth, James Lovelock, Oxford University Press, Oxford, United Kingdom, 1979.
                                  4. The Lion King, Walt Disney Pictures, Buena Vista Home Entertainment, 1994.
                                  5. “Solar and Economic Relationships,” Carlos Garcia-Mata and Felix Schaffner, Quarterly Journal of Economics, vol.
                                  49, no. 1, 1934, pp. 1-51.
                                  6. Curiously, 1929 was also the year that Hans Berger published the first recordings of human brain activity in his research
                                  attempting to understand the physiology of a youthful telepathic experience with his sister.
                                  7. “Declines of Biomes and Biotas and the Future of Evolution,” David S. Woodruff, Proceedings of the National Academy
                                  of Sciences of the United States of America, vol. 98, no. 10, 2001, pp. 5471-6.

                               24 | Annals of Improbable Research | July-August 2006                                                                                         www.improbable.com
Market Sector               mean Corr. Coeff.                n            % of total n
     Basic Materials                   0.03                       2                3%
     Consumer Goods                    0.23                       3                4%
     Financial                         0.31                       24              32%
     Healthcare                       -0.59                       10              14%
     Industrial Goods                 -0.19                       3                4%
     International                     0.83                       2                3%
     Services                         -0.41                       9               12%
     Technology                       -0.18                       16              22%
     Utilities                         0.72                       1                1%
     Not Specified                     0.37                       4                5%

  Table 1: Market Sectors and the mean Pearson’s correlation coefficients of responding stocks.

         HMO-NO News
         Health care advice to pass on to your patients

               Cure b
                                   y Comm
             We at H
            second
                        MO-NO            itment
                      to none are committed
            health c           .I                     to cure.
                     are orga t’s so strong th                 Our com
           program            nization              at we d               mitmen
                     . Sign u            offers: t          a re to of              t
           you wit             p fo r C           h e new H           f er some is
                   h1                   om                  MO                    thin
          least tw 00% commitm mitment Care -NO Commit g no other
                   o weeks               e                   ™, and               ment C
                            .* In the nt -- 24 hours                                        a
         further
                  services             unlikely            a day, s     w e g u arantee re™
                                                                   even da                t
                           , traditi
                                     onal no event that you                  ys a we o treat
                                             n-comm               re                   e k
                                                        itment p quire addition , for at
                                                                 rogram                a
                                                                          s are av l and/or
                                                                                    ailable.
                                                                                             **
                      * Note: Mayincur stochastic surcharge.
                      ** Note: On an as-available basis. Additional fees apply. May require a three-month prior notification and/or a six-
                      month membership re-consideration period.

           HMO-NO                                              The very final word in health care

www.improbable.com                                                                                 July-August 2006 | Annals of Improbable Research | 25
What is this picture?                                     (see page 1)      PERIODICALS

ISSN 1079-5146                                                           Annals of Improbable Research
Volume 12, Number 4                                                                     P.O. Box 380853
                                                                          www.improbable.com
July/August 2006                                                            Cambridge, MA 02238, USA
  34 | Annals of Improbable Research | July-August 2006                               (+1) 617.491.4437
                                                                                       www.improbable.com
You can also read